Dictionary Learning Based Near-Field Channel Estimation for Wideband XL-MIMO Systems

Extremely large-scale MIMO (XL-MIMO) is pivotal for enabling future 6G networks. However, accurate channel estimation for XL-MIMO faces a new challenge known as the near-field beam split effect. To capture the characteristics of the channel, this paper proposes a dictionary learning based channel estimation algorithm for wideband XL-MIMO systems. The channel estimation problem is formulated as a bilevel optimization task, where the upper-level aims to learn the optimal dictionary while the lower-level aims to find sparse channel representations based on the learned dictionary. To efficiently solve this bilevel optimization problem, we unfold the iterative shrinkage-thresholding algorithm into a deep neural network architecture. Learnable network blocks are employed to represent the dictionary, thus enabling adaptation to the characteristics of near-field wideband channel in XL-MIMO. The unfolded network is trained in an unsupervised manner by minimizing the reconstruction error via backpropagation and stochastic gradient descent. This eliminates the need for explicit gradient calculations associated with the bilevel optimization. Simulation results demonstrate that the proposed algorithm achieves good channel estimation performance, while offering lower computational complexity than the conventional compressed sensing based estimators.

I. Introduction

Extremely large-scale MIMO (XL-MIMO) employs antennas with a 10-fold increase in scale compared to conventional systems, thereby enabling a substantial enhancement of the spatial multiplexing gain. Furthermore, XL-MIMO’s ability to form extremely narrow beams with high directivity gain mitigates the severe path loss in millimeter wave (mmWave) and terahertz (THz) bands [1]. The integration of wideband high-frequency communications and XL-MIMO is thus regarded as a key enabler for future 6G wireless networks [2]. Acquiring accurate channel state information (CSI) is key to realizing the full performance gains of wideband XL-MIMO systems. However, the substantial number of antennas results in prohibitive pilot overhead for conventional channel estimation schemes.